In order to ensure safety and prevent collisions on road, automotive radars must be fault proof and have to be tested on reliability and performance, which requires proper diagnostic of well-functioning of the radar so that the car may participate to traffic. One way of the diagnostic of well-functioning of the automotive radar is by means of calibration in service stations. In contrast to offline calibration in service stations, one another way of testing the automotive radar on well-functioning would be by means of monitoring the state, or with other words the healthiness, of the radar in real-time. One possible solution to monitor the radar state is to use a massive set of calibration targets in road infrastructure and thereby the problem lies in optimally estimating the state based on the RCS information provided by the radar. As a result, in the first part of this thesis, based on the defined target selection criteria, selection of the most appropriate calibration target among possible candidates is discussed. Using massive set of targets is required to overcome the uncertainty in production and installation accuracy in one-target measurements. However, this method brings randomness which is caused by two error sources being the non-ideal shapes of the calibration targets originating from mass production errors and orientation errors either from installation or maintenance errors. Second part of the thesis investigates how this randomness affects the self-diagnostics performance of automotive radar. Thereby, the model of target orientation and RCS loss due to orientation errors, and, the model of target RCS and its RCS loss due to mass production errors are developed. For both error sources, the statistical characteristics of the loss factors are determined by means of corresponding probability distribution functions which are derived analytically. In case of orientation errors, analytical results are validated by Monte-Carlo simulations as well as Kullback-Leibler Divergence. In case of mass production errors, analytical results are validated by Monte-Carlo simulations only. Together with the results obtained in the first part, results of the statistical characteristics of the loss factor due to non-orthogonality help to find a balance between the size, quality and number of targets to be deployed in a certain range in a given road configuration. To finalize the project, the measurement model is determined according to which the approach for self-diagnostics is developed under certain assumptions and considering a set of measurements of the same realization of a single target only. By the developed approaches, the relation of statistical parameters on self-diagnostics performance is determined by means of three different estimation methods and the results are validated by Monte-Carlo simulations.